Search Engines in an AI Era: The False Promise of Factual and Verifiable Source-Cited Responses (Salesforce, arXiv 2410.22349)

Forum topic · 小凯 · 2026-07-05

Summary

This forum post indexes and annotates the October 2024 Salesforce research paper 'Search Engines in an AI Era: The False Promise of Factual and Verifiable Source-Cited Responses' (arXiv:2410.22349), authored by Pranav Narayanan Venkit, Philippe Laban, Yilun Zhou, Yixin Mao, and Chien-Sheng Wu. The paper examines whether answer engines built on retrieval-augmented generation (RAG) actually deliver on their promise of factual, verifiable, source-cited responses. The forum entry frames the work within the broader evolution of neural information retrieval—from BM25 and dense retrieval to generative retrieval and agentic search—and situates it in an RAG evaluation repository alongside related resources such as the RAG evaluation survey (arXiv:2405.07437), ARES, and Columbia Journalism Review's 'AI Search Has a Citation Problem.' The post includes engineering checklists covering data privacy, latency budgets, quality verification, security (corpus poisoning, bias), and cost control, plus a glossary of IR/RAG terms. It is aimed at researchers reproducing RAG evaluations, engineers integrating retrievers and rerankers, and product managers assessing answer trustworthiness beyond offline metrics like nDCG.

This post catalogs a Salesforce research paper from October 2024 that critically evaluates whether AI-powered answer engines—search systems built on retrieval-augmented generation (RAG)—truly provide factual and verifiable, source-cited responses as advertised.

Paper metadata

Title: Search Engines in an AI Era: The False Promise of Factual and Verifiable Source-Cited Responses
Authors / affiliation: Pranav Narayanan Venkit, Philippe Laban, Yilun Zhou, Yixin Mao, Chien-Sheng Wu (Salesforce)
Published: October 2024
Source: https://arxiv.org/abs/2410.22349
Type: Academic paper, indexed under the 'Evaluation of Search Engines' section of a RAG/answer-engine evaluation repository

Context within the field

The post situates the paper in the ongoing evolution of neural information retrieval: from BM25 and sparse retrieval, through dense bi-encoder retrieval and cross-encoder reranking, to generative retrieval and LLM-based agentic search. In the LLM era, key new variables include inference budget and action space—whether to retrieve, how many times, and which tools to call. RAG and agentic search extend external knowledge access from a one-shot retrieval event into an iterative, verifiable, plannable process, shifting evaluation from static metrics like nDCG toward task success rate, citation accuracy, and multi-hop reasoning-chain completeness.

Engineering checklist (from the post)

| Area | Concern | Recommendation | |------|---------|----------------| | Data | PII in training/index data, versioning | Partitioned indexes, anonymization, rollback-safe embedding versions | | Latency | p99 budget, number of retrieval steps | Cascades with early stopping, query caching, async reranking | | Quality | Does offline gain translate to online satisfaction? | Interleaving experiments, human audits, citation verification | | Safety | Poisoning/bias via open retrieval | Source whitelists, adversarial detection, output filtering | | Cost | Token and GPU cost per query | Small-model routing, distillation, hybrid sparse+dense retrieval |

Related entries cross-referenced in the repository

AI Search Has A Citation Problem (Columbia Journalism Review, Mar 2025)
Evaluation of Retrieval-Augmented Generation: A Survey (arXiv:2405.07437)
A Dataset of Information-Seeking Questions and Answers Anchored in Research Articles (arXiv:2105.03011)
ARES: An Automated Evaluation Framework for RAG (arXiv:2311.09476)
AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agents (arXiv:2401.13178)

Takeaways for readers

1. Researchers: Replicate the paper's comparison experiments; check whether statistical significance and compute cost are reported. 2. Engineers: Assess pluggable modules (encoders, rerankers, planners) and integration cost against existing stacks. 3. Product managers: Focus on user-perceivable benefits—latency, answer trustworthiness, multi-turn consistency—rather than offline nDCG alone.

> Note: The original forum post is largely a template-based catalog entry; quantitative findings from the paper should be verified against the PDF at https://arxiv.org/abs/2410.22349 before citation.